Registration of complex and self similar images such as plant canopy images is a challenge in plant sciences. Yet, this is often a required step for multimodal imaging, where unaligned sensors yield unregistered image pairs. We propose a pipeline adapted to such constraints, applied t o a pple tree canopies. Specifically, we apply an intensity based registration on downscaled and/or Gaussian blurred versions of the targeted images. This helps to eliminate spurious details, which smooths the optimization landscape and also helps to reduce differences between the modalities. Resul ts show better registration than with standard feature based or intensity based methods.